now that we have allowed for programmatic question specification, we potentially want to keep track of how questions are phrased (as a likely study involves asking multiple different variations on the same question to see which questions result in the highest accuracy)